Modify API for advanced settings. (SDCA) #2093

abgoswam · 2019-01-09T19:21:00Z

Towards #1798 .

This PR addresses only the SDCA agos. For the other remaining components, we will have separate PRs.

The following changes have been made:

Two public extension methods, one for simple arguments and the other for advanced options
Make constructors internal
Pass Options objects as arguments instead of Action delegate
Rename Arguments to Options
Rename Options objects as options (instead of args or advancedSettings used so far)
Added Help text as the XML comment of Options

abgoswam · 2019-01-14T16:00:10Z

test/Microsoft.ML.Tests/ScenariosWithDirectInstantiation/TensorflowTests.cs

@@ -25,7 +25,8 @@ private class TestData
            public float[] b;
        }

-        [ConditionalFact(typeof(Environment), nameof(Environment.Is64BitProcess))] // TensorFlow is 64-bit only
+        [Fact(Skip = "TF Tests fail")]


Please do not review this file. I am skipping TF tests only because they seem to be flaky at this point.

Comment here
#Closed

sfilipi · 2019-01-15T18:19:50Z

src/Microsoft.ML.StandardLearners/Standard/SdcaBinary.cs

@@ -69,6 +69,13 @@ public abstract class LinearTrainerBase<TTransformer, TModel> : TrainerEstimator
        {
        }

+        private protected LinearTrainerBase(IHostEnvironment env, string featureColumn, SchemaShape.Column labelColumn,
+            SchemaShape.Column weightColumn)


why do you need this? #Resolved

I added this to account for the corresponding changes in line #1728-1730

I saw some tests fail saying "Weights" column not found. The fix was to specify options.WeightColumn.IsExplicit in the call below.

TrainerUtils.MakeR4ScalarWeightColumn(options.WeightColumn, options.WeightColumn.IsExplicit)

In reply to: 248006972 [](ancestors = 248006972)

Got rid of this

In reply to: 248058147 [](ancestors = 248058147,248006972)

sfilipi · 2019-01-15T19:00:06Z

src/Microsoft.ML.StandardLearners/Standard/SdcaBinary.cs

@@ -1495,11 +1496,11 @@ internal override void Check(IHostEnvironment env)
            _outputColumns = outCols.ToArray();
        }

-        internal SdcaBinaryTrainer(IHostEnvironment env, Arguments args,
+        internal SdcaBinaryTrainer(IHostEnvironment env, Options options,
            string featureColumn, string labelColumn, string weightColumn = null)


string featureColumn, string labelColumn, string weightColumn = null [](start = 11, length = 69)

are those still needed? can they be passed to options? The other ctor below can go away than. #Resolved

Fixed.

In reply to: 248021483 [](ancestors = 248021483)

sfilipi · 2019-01-15T19:01:12Z

src/Microsoft.ML.StandardLearners/Standard/SdcaBinary.cs

@@ -1594,7 +1595,7 @@ public sealed class StochasticGradientDescentClassificationTrainer :
        internal const string UserNameValue = "Hogwild SGD (binary)";
        internal const string ShortName = "HogwildSGD";

-        public sealed class Arguments : LearnerInputBaseWithWeight
+        public sealed class Options : LearnerInputBaseWithWeight
        {
            [Argument(ArgumentType.Multiple, HelpText = "Loss Function", ShortName = "loss", SortOrder = 50)]
            public ISupportClassificationLossFactory LossFunction = new LogLossFactory();


ISupportClassificationLossFactory [](start = 19, length = 33)

XML docs #Resolved

Created #2154 to fix this issue in bulk for several learners

In reply to: 248022036 [](ancestors = 248022036)

sfilipi · 2019-01-15T21:00:19Z

    /// </example>

please remove duplicate example #Resolved

Refers to: src/Microsoft.ML.StandardLearners/StandardLearnersCatalog.cs:116 in f0b9565. [](commit_id = f0b9565, deletion_comment = False)

sfilipi · 2019-01-15T21:00:37Z

src/Microsoft.ML.StandardLearners/StandardLearnersCatalog.cs

+        /// <![CDATA[
+        ///  [!code-csharp[SDCA](~/../docs/samples/docs/samples/Microsoft.ML.Samples/Dynamic/SDCA.cs)]
+        /// ]]></format>
+        /// </example>


please remove duplicate example #Resolved

artidoro · 2019-01-15T23:08:19Z

src/Microsoft.ML.StandardLearners/StandardLearnersCatalog.cs

+            Contracts.CheckValue(ctx, nameof(ctx));
+            var env = CatalogUtils.GetEnvironment(ctx);
+
+            return new StochasticGradientDescentClassificationTrainer(env, options);


Minor comment: everywhere else you don't skip the line. #Resolved

artidoro · 2019-01-15T23:13:13Z

src/Microsoft.ML.StandardLearners/Standard/SdcaBinary.cs

            : base(env, featureColumn, TrainerUtils.MakeBoolScalarLabel(labelColumn), weightColumn)
        {
            Host.CheckNonEmpty(featureColumn, nameof(featureColumn));
            Host.CheckNonEmpty(labelColumn, nameof(labelColumn));

-            _args = new Arguments();
+            _args = new Options();


_args [](start = 12, length = 5)

Could you make this _options? So that it is consistent with the general renaming? #Resolved

artidoro · 2019-01-15T23:27:48Z

src/Microsoft.ML.StandardLearners/Standard/SdcaBinary.cs

+        /// <param name="env">The environment to use.</param>
+        /// <param name="options">Advanced arguments to the algorithm.</param>
+        internal StochasticGradientDescentClassificationTrainer(IHostEnvironment env, Options options)
+            : base(env, options.FeatureColumn, TrainerUtils.MakeBoolScalarLabel(options.LabelColumn), TrainerUtils.MakeR4ScalarWeightColumn(options.WeightColumn, options.WeightColumn.IsExplicit))


TrainerUtils.MakeR4ScalarWeightColumn(options.WeightColumn, options.WeightColumn.IsExplicit) [](start = 102, length = 92)

I think you can replace this with:

options.WeightColumn.IsExplicit ? options.WeightColumn : null

That way you won't need the other constructor for the base. This is a strange problem from wanting to keep the optional for the weightColumn which is used in Maml. #Resolved

@sfilipi who is taking a look at how best to handle the weights column

In this PR, I would like to keep it as is. This has ramifications for several tests.

In reply to: 248102195 [](ancestors = 248102195)

Should be a pretty simple fix, which reduces the number of constructors in the base class, I would check if it works :)

In reply to: 248104112 [](ancestors = 248104112,248102195)

Modified. Thanks for debugging !

In reply to: 248104779 [](ancestors = 248104779,248104112,248102195)

artidoro · 2019-01-15T23:30:15Z

src/Microsoft.ML.StandardLearners/Standard/SdcaMultiClass.cs

    {
        public const string LoadNameValue = "SDCAMC";
        public const string UserNameValue = "Fast Linear Multi-class Classification (SA-SDCA)";
        public const string ShortName = "sasdcamc";
        internal const string Summary = "The SDCA linear multi-class classification trainer.";

-        public sealed class Arguments : ArgumentsBase
+        public sealed class Options : ArgumentsBase


ArgumentsBase [](start = 38, length = 13)

Are we planning to rename this too? #Resolved

At this point the focus is to get the public APIs fixed.

Rename of internal classes / variables is low priority at this point.

In reply to: 248102673 [](ancestors = 248102673)

artidoro · 2019-01-15T23:36:47Z

test/Microsoft.ML.StaticPipelineTesting/Training.cs

@@ -48,7 +48,7 @@ public void SdcaRegression()

            var est = reader.MakeNewEstimator()
                .Append(r => (r.label, score: ctx.Trainers.Sdca(r.label, r.features, maxIterations: 2,
-                onFit: p => pred = p, advancedSettings: s => s.NumThreads = 1)));


NumThreads [](start = 63, length = 10)

Would be good to have numthreads to 1.
Unfortunately, when that's not the case, it is more prone to having random failures from small discrepancies in test results especially on 32 bit machines. #Resolved

fixed

In reply to: 248104144 [](ancestors = 248104144)

artidoro · 2019-01-15T23:37:03Z

test/Microsoft.ML.StaticPipelineTesting/Training.cs

@@ -88,7 +88,7 @@ public void SdcaRegressionNameCollision()
                separator: ';', hasHeader: true);

            var est = reader.MakeNewEstimator()
-                .Append(r => (r.label, r.Score, score: ctx.Trainers.Sdca(r.label, r.features, maxIterations: 2, advancedSettings: s => s.NumThreads = 1)));


.NumThreads = 1 [](start = 136, length = 15)

Same here. #Resolved

fixed

In reply to: 248104199 [](ancestors = 248104199)

artidoro · 2019-01-15T23:37:17Z

test/Microsoft.ML.StaticPipelineTesting/Training.cs

@@ -121,8 +121,7 @@ public void SdcaBinaryClassification()
            var est = reader.MakeNewEstimator()
                .Append(r => (r.label, preds: ctx.Trainers.Sdca(r.label, r.features,
                    maxIterations: 2,
-                    onFit: (p, c) => { pred = p; cali = c; },
-                    advancedSettings: s => s.NumThreads = 1)));


s.NumThreads = 1 [](start = 42, length = 17)

And here. #Resolved

fixed

In reply to: 248104248 [](ancestors = 248104248)

artidoro · 2019-01-15T23:40:23Z

test/Microsoft.ML.StaticPipelineTesting/Training.cs

@@ -169,8 +168,7 @@ public void SdcaBinaryClassificationNoCalibration()
            var est = reader.MakeNewEstimator()
                .Append(r => (r.label, preds: ctx.Trainers.Sdca(r.label, r.features,
                maxIterations: 2,
-                loss: loss, onFit: p => pred = p,
-                advancedSettings: s => s.NumThreads = 1)));


s.NumThreads = 1 [](start = 38, length = 17)

Here too. #Resolved

fixed

In reply to: 248104923 [](ancestors = 248104923)

artidoro · 2019-01-15T23:40:34Z

test/Microsoft.ML.StaticPipelineTesting/Training.cs

@@ -940,8 +938,7 @@ public void HogwildSGDBinaryClassification()
            var est = reader.MakeNewEstimator()
                .Append(r => (r.label, preds: ctx.Trainers.StochasticGradientDescentClassificationTrainer(r.label, r.features,
                    l2Weight: 0,
-                    onFit: (p) => { pred = p; },
-                    advancedSettings: s => s.NumThreads = 1)));


s.NumThreads = 1 [](start = 43, length = 16)

And here. #Resolved

sfilipi · 2019-01-17T17:34:39Z

src/Microsoft.ML.StaticPipe/SdcaStaticExtensions.cs

+            Contracts.CheckValueOrNull(onFit);
+
+            var rec = new TrainerEstimatorReconciler.Regression(
+                (env, labelName, featuresName, weightsName) =>


weightsName [](start = 47, length = 11)

this is not getting used. #Pending

Created issue #2175 to look into usages of weights parameter in our public API

For several learners (e.g. SDCA) it seems not necessary

In reply to: 248769861 [](ancestors = 248769861)

sfilipi · 2019-01-17T17:35:04Z

src/Microsoft.ML.StandardLearners/Standard/SdcaRegression.cs

-        /// <param name="advancedSettings">A delegate to set more settings.
-        /// The settings here will override the ones provided in the direct method signature,
-        /// if both are present and have different values.
-        /// The columns names, however need to be provided directly, not through the <paramref name="advancedSettings"/>.</param>
        public SdcaRegressionTrainer(IHostEnvironment env,


public SdcaRegressionTrainer(IHos [](start = 5, length = 36)

shall this ctor go away? #Resolved

made internal. deletion will happen in #2100

In reply to: 248770050 [](ancestors = 248770050)

sfilipi · 2019-01-17T17:35:51Z

src/Microsoft.ML.StandardLearners/Standard/SdcaRegression.cs

        {
            Host.CheckNonEmpty(featureColumn, nameof(featureColumn));
            Host.CheckNonEmpty(labelColumn, nameof(labelColumn));
            _loss = loss ?? Args.LossFunction.CreateComponent(env);
            Loss = _loss;
        }

-        internal SdcaRegressionTrainer(IHostEnvironment env, Arguments args, string featureColumn, string labelColumn, string weightColumn = null)
-            : base(env, args, TrainerUtils.MakeR4ScalarColumn(labelColumn), TrainerUtils.MakeR4ScalarWeightColumn(weightColumn))
+        internal SdcaRegressionTrainer(IHostEnvironment env, Options options, string featureColumn, string labelColumn, string weightColumn = null)


internal SdcaRegressionTrainer [](start = 7, length = 31)

I'd remove this one too. #Resolved

#2100

In reply to: 248770295 [](ancestors = 248770295)

sfilipi · 2019-01-17T17:36:12Z

src/Microsoft.ML.StandardLearners/Standard/SdcaMultiClass.cs

        {
            Host.CheckNonEmpty(featureColumn, nameof(featureColumn));
            Host.CheckNonEmpty(labelColumn, nameof(labelColumn));
            _loss = loss ?? Args.LossFunction.CreateComponent(env);
            Loss = _loss;
        }

-        internal SdcaMultiClassTrainer(IHostEnvironment env, Arguments args,
+        internal SdcaMultiClassTrainer(IHostEnvironment env, Options options,


internal SdcaMultiClassTrainer [](start = 8, length = 30)

delete, maybe #Resolved

#2100

In reply to: 248770420 [](ancestors = 248770420)

sfilipi · 2019-01-17T17:36:19Z

src/Microsoft.ML.StandardLearners/Standard/SdcaMultiClass.cs

-        /// <param name="advancedSettings">A delegate to set more settings.
-        /// The settings here will override the ones provided in the direct method signature,
-        /// if both are present and have different values.
-        /// The columns names, however need to be provided directly, not through the <paramref name="advancedSettings"/>.</param>
        public SdcaMultiClassTrainer(IHostEnvironment env,


public SdcaMultiClassTrainer [](start = 6, length = 30)

same #Resolved

made internal. deletion will happen in #2100

In reply to: 248770465 [](ancestors = 248770465)

sfilipi · 2019-01-17T17:37:27Z

src/Microsoft.ML.StandardLearners/Standard/SdcaBinary.cs

        {
            var args = new TArgs();

-            // Apply the advanced args, if the user supplied any.
-            advancedSettings?.Invoke(args);
            args.FeatureColumn = featureColumn;
            args.LabelColumn = labelColumn.Name;
            return args;
        }

        internal SdcaTrainerBase(IHostEnvironment env, string featureColumn, SchemaShape.Column labelColumn,


internal SdcaTrainerBase [](start = 6, length = 26)

delete #Resolved

#2100

In reply to: 248770895 [](ancestors = 248770895)

sfilipi · 2019-01-17T17:37:53Z

src/Microsoft.ML.StandardLearners/Standard/SdcaBinary.cs

@@ -1391,12 +1389,13 @@ public void Add(Double summand)
        }
    }

-    public sealed class SdcaBinaryTrainer : SdcaTrainerBase<SdcaBinaryTrainer.Arguments, BinaryPredictionTransformer<TScalarPredictor>, TScalarPredictor>
+    public sealed class SdcaBinaryTrainer : SdcaTrainerBase<SdcaBinaryTrainer.Options, BinaryPredictionTransformer<TScalarPredictor>, TScalarPredictor>
    {
        public const string LoadNameValue = "SDCA";


public [](start = 8, length = 6)

all those strings should be internal.. #Resolved

is there some separate issue for it that I can link to this comment ?

I would like to keep the PRs for public API focused on API rather than fixing issues across the codebase :)

In reply to: 248771027 [](ancestors = 248771027)

sfilipi · 2019-01-17T17:40:18Z

src/Microsoft.ML.StandardLearners/Standard/SdcaBinary.cs

-
-        }
-
-        public SdcaBinaryTrainer(IHostEnvironment env, Arguments args)


public SdcaBinaryTrainer(IHostEnvironment env, Arguments args) [](start = 6, length = 64)

i think this is needed for the SignatureTrainer. Doesn't have to be public. #Resolved

the constructor above has this signature. so its not really going away.

In reply to: 248771919 [](ancestors = 248771919)

sfilipi · 2019-01-17T17:42:15Z

src/Microsoft.ML.StandardLearners/Standard/SdcaBinary.cs

@@ -1688,53 +1675,50 @@ internal static class Defaults
        /// <param name="initLearningRate">The initial learning rate used by SGD.</param>
        /// <param name="l2Weight">The L2 regularizer constant.</param>
        /// <param name="loss">The loss function to use.</param>
-        /// <param name="advancedSettings">A delegate to apply all the advanced arguments to the algorithm.</param>
-        public StochasticGradientDescentClassificationTrainer(IHostEnvironment env,
+        internal StochasticGradientDescentClassificationTrainer(IHostEnvironment env,


internal StochasticGradientDescentClassificationTraine [](start = 8, length = 54)

I'd remove all ctors but the one with the (IHostEnvironment env, Arguments) signature, unless is needed for inheritance. #Resolved

exactly. thats the point of issue #2100. but that would be a separate PR.

these PRs focus on public API which is P0 (as per consensus in email thread)

In reply to: 248772619 [](ancestors = 248772619)

singlis · 2019-01-17T22:27:16Z

src/Microsoft.ML.StandardLearners/Standard/SdcaRegression.cs

            Loss = _loss;
        }

-        internal SdcaRegressionTrainer(IHostEnvironment env, Arguments args)
-            : this(env, args, args.FeatureColumn, args.LabelColumn)
+        internal SdcaRegressionTrainer(IHostEnvironment env, Options options)


SdcaRegressionTrainer [](start = 17, length = 21)

isnt there suposed to be a constructor with common arguments and one with options? If we make the other two constructors internal (and eventually remove), then we have just this constructor with the options. Is that OK? Does SDCARegressionTrainer have any common args? #Pending

not quite sure what you mean.

After #2100, only 1 constructor should remain.

In reply to: 248865494 [](ancestors = 248865494)

I thought that main point with the API changes is that we end up with two constructors that are publicly available to the user:

one constructor that would be used for most scenarios and contains a set of arguments that are just passed in Foo(arg1, arg2...)

one constructor that only takes in options if we want to set advanced features Foo(options)

Now the first constructor can use the options constructor under the covers, i.e. Foo(arg1, arg2): this(new Options(){ arg1 = arg1, arg2 = arg2 }... but we would still expose two different constructors publicly that can be used.

Is that not the case for SDCA?

In reply to: 248870888 [](ancestors = 248870888,248865494)

We should have two MLContext extensions that are publicly available to the user. This is the public API for ML.NET

All constructors are internal

With Refactoring of Constructors #2100 we will unify the internal constructors. But thats P2 for now.

In reply to: 248880304 [](ancestors = 248880304,248870888,248865494)

singlis · 2019-01-17T22:30:35Z

src/Microsoft.ML.StandardLearners/Standard/SdcaBinary.cs

-        internal SdcaBinaryTrainer(IHostEnvironment env, Arguments args,
-            string featureColumn, string labelColumn, string weightColumn = null)
-            : base(env, args, TrainerUtils.MakeBoolScalarLabel(labelColumn), TrainerUtils.MakeR4ScalarWeightColumn(weightColumn))
+        internal SdcaBinaryTrainer(IHostEnvironment env, Options options)


Same comment here as I made with the SDCARegression -- does SdcaBinary not have common args or is it always options? #Resolved

we will eventually have just 1 constructor (the one with Options)

#2100

In reply to: 248866446 [](ancestors = 248866446)

singlis · 2019-01-17T22:34:10Z

src/Microsoft.ML.StandardLearners/Standard/SdcaBinary.cs

@@ -1688,53 +1674,50 @@ internal static class Defaults
        /// <param name="initLearningRate">The initial learning rate used by SGD.</param>
        /// <param name="l2Weight">The L2 regularizer constant.</param>
        /// <param name="loss">The loss function to use.</param>
-        /// <param name="advancedSettings">A delegate to apply all the advanced arguments to the algorithm.</param>
-        public StochasticGradientDescentClassificationTrainer(IHostEnvironment env,
+        internal StochasticGradientDescentClassificationTrainer(IHostEnvironment env,
            string labelColumn = DefaultColumnNames.Label,
            string featureColumn = DefaultColumnNames.Features,
            string weightColumn = null,


Can this be set to Defaults.WeightColumn? #Resolved

we need to take closer look at how 'weights' are being used, or if they are being used at all

please see #2175.

In reply to: 248867308 [](ancestors = 248867308)

singlis · 2019-01-17T22:50:40Z

src/Microsoft.ML.StandardLearners/Standard/SdcaBinary.cs

+        /// <param name="env">The environment to use.</param>
+        /// <param name="options">Advanced arguments to the algorithm.</param>
+        internal StochasticGradientDescentClassificationTrainer(IHostEnvironment env, Options options)
+            : base(env, options.FeatureColumn, TrainerUtils.MakeBoolScalarLabel(options.LabelColumn), options.WeightColumn.IsExplicit ? options.WeightColumn.Value : null)


I am confused on why the if check is needed. You are saying that if the WeightColumn is explicitly set use the value, otherwise pass null. If null is used when its not set, then is DefaultColumnNames.Weight needed? If not, could the line above:
_options.WeightColumn = ... : Optional.Implicit(DefaultColumnNames.Weight) to:
_options.WeightColumn = ... : Optional.Implicit(null)

To me the code above does the branching to figure out what parameter to use - its weird to branch afterwards to figure out if its explicit or not.
#Resolved

couldn't follow the question properly .

basically without this, some tests fail. also i created #2175 to verify if weights is actually being used at all or not

In reply to: 248871449 [](ancestors = 248871449)

I see what is happening. So Options.WeightColumn is an Optional and it does default to DefaultColumNames.Weight...
Maybe change this line:
_options.WeightColumn = weightColumn != null ? Optional.Explicit(weightColumn) : Optional.Implicit(DefaultColumnNames.Weight);
to
_options.WeightColumn = weightColumn != null ? Optional.Explicit(weightColumn) ;

Since its setup with a default?

In reply to: 248873304 [](ancestors = 248873304,248871449)

singlis · 2019-01-17T22:54:29Z

src/Microsoft.ML.StaticPipe/SdcaStaticExtensions.cs

+        }
+
+        /// <summary>
+        /// Predict a target using a linear binary classification model trained with the SDCA trainer, and log-loss.


minor, you can remove the comma for , and log-loss.

singlis · 2019-01-17T22:55:21Z

src/Microsoft.ML.StaticPipe/SdcaStaticExtensions.cs

+        /// Predict a target using a linear binary classification model trained with the SDCA trainer, and log-loss.
+        /// </summary>
+        /// <param name="ctx">The binary classification context trainer object.</param>
+        /// <param name="label">The label, or dependent variable.</param>


same here: label or dependent variable. Same thing with the line below "features or independent weights". Same comment applies to other apis as it looks like this comment was copied.

singlis · 2019-01-17T22:55:38Z

src/Microsoft.ML.StaticPipe/SdcaStaticExtensions.cs

+        /// <param name="label">The label, or dependent variable.</param>
+        /// <param name="features">The features, or independent variables.</param>
+        /// <param name="weights">The optional example weights.</param>
+        /// <param name="options">Advanced arguments to the algorithm.</param>


arguments [](start = 43, length = 9)

options? #Pending

#2154

In reply to: 248872745 [](ancestors = 248872745)

same comment here -- this is all new code, why not change it here?

In reply to: 248872978 [](ancestors = 248872978,248872745)

singlis · 2019-01-17T22:57:27Z

src/Microsoft.ML.StaticPipe/SdcaStaticExtensions.cs

+                    SdcaBinaryTrainer.Options options,
+                    Action<LinearBinaryModelParameters, ParameterMixingCalibratedPredictor> onFit = null)
+        {
+            Contracts.CheckValue(label, nameof(label));


Just curious - this is a documented example right? Is it worth having CheckValues in the example? #Resolved

this is the Static API extension

In reply to: 248873167 [](ancestors = 248873167)

singlis · 2019-01-17T23:00:40Z

src/Microsoft.ML.StaticPipe/SdcaStaticExtensions.cs

+        /// from negative to positive infinity), and the predicted label.</returns>
+        public static (Scalar<float> score, Scalar<bool> predictedLabel) Sdca(
+                this BinaryClassificationContext.BinaryClassificationTrainers ctx,
+                Scalar<bool> label, Vector<float> features,


Vector [](start = 36, length = 6)

Should features go on a separate line to be consistent with the other args?

singlis · 2019-01-17T23:02:39Z

src/Microsoft.ML.StaticPipe/SgdStatic.cs

+        /// <param name="label">The name of the label column.</param>
+        /// <param name="features">The name of the feature column.</param>
+        /// <param name="weights">The name for the example weight column.</param>
+        /// <param name="options">Advanced arguments to the algorithm.</param>


arguments [](start = 43, length = 9)

options? #Pending

#2154

In reply to: 248874445 [](ancestors = 248874445)

But this is a new comment/new code. why not update it with this PR?

In reply to: 248874778 [](ancestors = 248874778,248874445)

artidoro

singlis · 2019-01-17T23:04:00Z

test/Microsoft.ML.Core.Tests/UnitTests/TestEntryPoints.cs

@@ -2488,7 +2488,7 @@ public void TestInputBuilderComponentFactories()
            Assert.True(success);
            var inputBuilder = new InputBuilder(Env, info.InputType, catalog);

-            var args = new SdcaBinaryTrainer.Arguments()
+            var args = new SdcaBinaryTrainer.Options()


args [](start = 16, length = 4)

rename args to options?

singlis · 2019-01-17T23:06:26Z

test/Microsoft.ML.StaticPipelineTesting/Training.cs

-                .Append(r => (r.label, score: ctx.Trainers.Sdca(r.label, r.features, maxIterations: 2,
-                onFit: p => pred = p, advancedSettings: s => s.NumThreads = 1)));
+                .Append(r => (r.label, score: ctx.Trainers.Sdca(r.label, r.features, null,
+                new SdcaRegressionTrainer.Options() { MaxIterations = 2, NumThreads = 1 },


NumThreads [](start = 73, length = 10)

@ if numthreads should be 1, would it better to default NumThreads in options to 1? #Resolved

modifying default settings is out of purview of this PR.

i think we are using NumThreads = 1 only for test cases to be more stable

In reply to: 248875274 [](ancestors = 248875274)

singlis

singlis · 2019-01-17T23:42:22Z

Overall looks good - I left some minor feedback.

abgoswam added 4 commits January 9, 2019 19:20

dummy change to test build

1a58429

Merge branch 'master' into abgoswam/action_arguments_3

60eb9d5

StochasticGradientDescentClassificationTrainer

2a77e5e

SdcaBinaryTrainer, SdcaMultiClassTrainer, SdcaRegressionTrainer

f0b9565

abgoswam changed the title ~~WIP : Modify API for advanced settings. (SDCA)~~ Modify API for advanced settings. (SDCA) Jan 14, 2019

abgoswam requested review from sfilipi, artidoro and TomFinley January 14, 2019 03:44

abgoswam commented Jan 14, 2019

View reviewed changes

sfilipi reviewed Jan 15, 2019

View reviewed changes

review comments

192cfeb

artidoro reviewed Jan 15, 2019

View reviewed changes

review comments

0f9854e

artidoro reviewed Jan 15, 2019

View reviewed changes

abgoswam mentioned this pull request Jan 16, 2019

KMeans and Implicit weight cleanup #2158

Merged

abgoswam added 4 commits January 16, 2019 01:48

review comments

fd1cdf6

Merge branch 'master' into abgoswam/action_arguments_3

72c7d81

re-enable the TF tests + add more checks for options argument

86b7a35

Merge branch 'master' into abgoswam/action_arguments_3

0e10e0b

abgoswam added 2 commits January 16, 2019 20:56

merge with latest master to pick up overload of

5e4377b

review comments

feaaa65

sfilipi reviewed Jan 17, 2019

View reviewed changes

review comments

5fca29e

singlis reviewed Jan 17, 2019

View reviewed changes

artidoro approved these changes Jan 17, 2019

View reviewed changes

singlis reviewed Jan 17, 2019

View reviewed changes

singlis approved these changes Jan 17, 2019

View reviewed changes

abgoswam merged commit b65d725 into dotnet:master Jan 18, 2019

abgoswam deleted the abgoswam/action_arguments_3 branch January 21, 2019 21:25

ghost locked as resolved and limited conversation to collaborators Mar 25, 2022

Modify API for advanced settings. (SDCA) #2093

Modify API for advanced settings. (SDCA) #2093

Conversation

abgoswam commented Jan 9, 2019 • edited Loading

abgoswam Jan 14, 2019 • edited Loading

Choose a reason for hiding this comment

sfilipi Jan 15, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

abgoswam Jan 15, 2019 • edited Loading

Choose a reason for hiding this comment

sfilipi Jan 15, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

abgoswam Jan 15, 2019 • edited Loading

Choose a reason for hiding this comment

sfilipi Jan 15, 2019 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sfilipi commented Jan 15, 2019 • edited by abgoswam Loading

sfilipi Jan 15, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

artidoro Jan 15, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

artidoro Jan 15, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

artidoro Jan 15, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

abgoswam Jan 15, 2019 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

artidoro Jan 15, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

artidoro Jan 15, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

artidoro Jan 15, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

artidoro Jan 15, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

artidoro Jan 15, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

artidoro Jan 15, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

sfilipi Jan 17, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sfilipi Jan 17, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

abgoswam Jan 17, 2019 • edited Loading

Choose a reason for hiding this comment

sfilipi Jan 17, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sfilipi Jan 17, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sfilipi Jan 17, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sfilipi Jan 17, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sfilipi Jan 17, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sfilipi Jan 17, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sfilipi Jan 17, 2019 • edited by abgoswam Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

singlis Jan 17, 2019 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

abgoswam commented Jan 9, 2019 •

edited

Loading

abgoswam Jan 14, 2019 •

edited

Loading

sfilipi Jan 15, 2019 •

edited by abgoswam

Loading

abgoswam Jan 15, 2019 •

edited

Loading

sfilipi Jan 15, 2019 •

edited by abgoswam

Loading

abgoswam Jan 15, 2019 •

edited

Loading

sfilipi Jan 15, 2019 •

edited

Loading

sfilipi commented Jan 15, 2019 •

edited by abgoswam

Loading

sfilipi Jan 15, 2019 •

edited by abgoswam

Loading

artidoro Jan 15, 2019 •

edited by abgoswam

Loading

artidoro Jan 15, 2019 •

edited by abgoswam

Loading

artidoro Jan 15, 2019 •

edited by abgoswam

Loading

abgoswam Jan 15, 2019 •

edited

Loading

artidoro Jan 15, 2019 •

edited by abgoswam

Loading

artidoro Jan 15, 2019 •

edited by abgoswam

Loading

artidoro Jan 15, 2019 •

edited by abgoswam

Loading

artidoro Jan 15, 2019 •

edited by abgoswam

Loading

artidoro Jan 15, 2019 •

edited by abgoswam

Loading

artidoro Jan 15, 2019 •

edited by abgoswam

Loading

sfilipi Jan 17, 2019 •

edited by abgoswam

Loading

sfilipi Jan 17, 2019 •

edited by abgoswam

Loading

abgoswam Jan 17, 2019 •

edited

Loading

sfilipi Jan 17, 2019 •

edited by abgoswam

Loading

sfilipi Jan 17, 2019 •

edited by abgoswam

Loading

sfilipi Jan 17, 2019 •

edited by abgoswam

Loading

sfilipi Jan 17, 2019 •

edited by abgoswam

Loading

sfilipi Jan 17, 2019 •

edited by abgoswam

Loading

sfilipi Jan 17, 2019 •

edited by abgoswam

Loading

sfilipi Jan 17, 2019 •

edited by abgoswam

Loading

singlis Jan 17, 2019 •

edited

Loading

abgoswam Jan 18, 2019 •

edited

Loading

singlis Jan 17, 2019 •

edited by abgoswam

Loading

singlis Jan 17, 2019 •

edited by abgoswam

Loading

singlis Jan 17, 2019 •

edited by abgoswam

Loading

abgoswam Jan 17, 2019 •

edited

Loading

singlis Jan 17, 2019 •

edited

Loading

singlis Jan 17, 2019 •

edited

Loading

singlis Jan 17, 2019 •

edited by abgoswam

Loading

singlis Jan 17, 2019 •

edited

Loading

singlis Jan 17, 2019 •

edited by abgoswam

Loading